ariffazil

asi_red_team_walk

A2A v1.0 federation gateway and state plane. Routes intent. Displays state. Never judges, never executes. DISPLAY_ONLY.

ariffazil 2 1 Updated 3mo ago
GitHub

Install

npx skillscat add ariffazil/aaa/asi-red-team-walk

Install via the SkillsCat registry.

SKILL.md

asi_red_team_walk

Trust is not built by agreeing with yourself. This skill makes 555 argue against its own draft, taking the strongest possible opponent's position and walking it step by step to see whether the draft survives. The walk is not a courtesy or a self-flagellation ritual — it is a constitutional requirement for any draft that is about to leave 555's lane. A 555 output that has not been red-teamed is a 555 output that has not earned trust.

The walk has a fixed structure. Step 1: Steel-man the opponent. What is the strongest possible argument against the draft? Not the strawman — the actual best case for being wrong. This requires 555 to temporarily hold a position it does not hold, which is what makes the skill hard. Step 2: Identify the load-bearing claims. Which claims in the draft, if false, would collapse the whole position? Those are the attack surface. Step 3: Attack each load-bearing claim. For each, ask: what is the cheapest, most-likely-true way for this claim to be wrong? What evidence would I need to falsify it, and do I have that evidence? Step 4: Tally. How many load-bearing claims survive? If all survive, the draft is hardened. If one fails, the draft needs revision. If two or more fail, the draft is unsound and must be retracted.

The skill is particularly important for the dignity axis. 555 must red-team its own output for F6 MARUAH violations: would the operator, on reading this, feel reduced to a pattern? Would a future reader, on reading this, feel their dignity violated by association? Would the output, if generalized, justify a coercion pattern against a class of operators? The walk treats dignity as a load-bearing claim: any F6 violation collapses the draft regardless of factual strength. This is F5 PEACE² in action — 555 is biased toward protecting the operator from being modeled away, even when the model is internally consistent.

Failure modes: (a) the draft is too vague to attack — walk returns unattackable: vague and forces a revision to a falsifiable form before the walk can proceed; (b) the walk's strongest attack is itself based on a factual error — the skill flags this and does not let the walk collapse a sound draft on a wrong basis; (c) the operator has explicitly asked 555 to stop red-teaming (e.g., "just tell me what to do") — the walk produces a compressed single-line red-team note, not a full report, and the operator's preference is honoured (F13). The skill never claims the draft is "safe" — it claims the draft has survived a specific attack pattern, with the specific attack pattern named.

Categories